failproofai 1.0.12-beta.1 → 1.0.12

package/.next/standalone/fp-cloud-cli/tests/test_staging_parity.py

@@ -0,0 +1,198 @@
+"""Commands aligned with the FailproofAI Cloud staging API after the #885 port:
+every-kind policy backtests, Jev/both authoring, issue board filters and facets,
+evaluation facets, session import and judge/Jev backfill."""
+
+import gzip
+import json
+
+import pytest
+import respx
+
+from fp_cli.app import app
+
+BASE = "http://dash.test"
+CHECKS = [{"name": "prod-db", "question": "Does this touch the production database?"}]
+
+
+def sse(events):
+    return "".join(f"data: {json.dumps(e)}\n\n" for e in events)
+
+
+def test_backtest_regex_file_keeps_the_module_route(logged_in, runner, tmp_path):
+    src = tmp_path / "p.mjs"
+    src.write_text("export default {}")
+    with respx.mock as mock:
+        route = mock.post(f"{BASE}/api/agent/policy-backtest").respond(200, json={"matched": 0})
+        result = runner.invoke(app, ["--json", "policies", "backtest", str(src), "--agent-id", "a1", "--hours", "24"])
+    assert result.exit_code == 0, result.output
+    assert json.loads(route.calls[0].request.content) == {"source": "export default {}", "agentId": "a1", "hours": 24}
+
+
+def test_backtest_published_jev_policy_uses_backtest_all_estimate(logged_in, runner):
+    with respx.mock as mock:
+        mock.get(f"{BASE}/api/enforcement/policies").respond(200, json=[
+            {"id": "p1", "version": 1, "kind": "jev", "semantic": CHECKS},
+        ])
+        route = mock.post(f"{BASE}/api/agent/policy-backtest-all").respond(200, json={"kind": "jev", "mode": "estimate"})
+        result = runner.invoke(app, ["--json", "policies", "backtest", "p1"])
+    assert result.exit_code == 0, result.output
+    assert json.loads(route.calls[0].request.content) == {
+        "kind": "jev", "semantic": CHECKS, "hours": 168, "mode": "estimate",
+    }
+
+
+def test_backtest_both_from_files_with_run(logged_in, runner, tmp_path):
+    src, sem = tmp_path / "p.mjs", tmp_path / "checks.json"
+    src.write_text("export default {}")
+    sem.write_text(json.dumps(CHECKS))
+    with respx.mock as mock:
+        route = mock.post(f"{BASE}/api/agent/policy-backtest-all").respond(200, json={"kind": "both"})
+        result = runner.invoke(app, ["--json", "policies", "backtest", str(src), "--semantic", str(sem),
+                                     "--run", "--sample", "10"])
+    assert result.exit_code == 0, result.output
+    body = json.loads(route.calls[0].request.content)
+    assert body["kind"] == "both" and body["mode"] == "run" and body["sample"] == 10
+    assert body["source"] == "export default {}" and body["semantic"] == CHECKS
+
+
+@pytest.mark.parametrize("argv", [
+    ["policies", "backtest"],
+    ["policies", "backtest", "--kind", "jev", "--semantic", "missing.json"],
+    ["policies", "backtest", "x.mjs", "--kind", "nope"],
+    ["policies", "backtest", "x.mjs", "--sample", "99"],
+])
+def test_backtest_usage_errors(logged_in, runner, argv):
+    with respx.mock(assert_all_called=False):
+        result = runner.invoke(app, ["--json", *argv])
+    assert result.exit_code == 2, result.output
+
+
+@pytest.mark.parametrize("kind,sent", [(None, None), ("regex", None), ("jev", "jev"), ("both", "both")])
+def test_author_kind(logged_in, runner, kind, sent):
+    with respx.mock as mock:
+        route = mock.post(f"{BASE}/api/agent/policy-author").respond(
+            200, text=sse([{"type": "done", "result": {"source": "x"}}]))
+        result = runner.invoke(app, ["--json", "policies", "author", "No prod DB",
+                                     *(["--kind", kind] if kind else [])])
+    assert result.exit_code == 0, result.output
+    assert json.loads(route.calls[0].request.content).get("kind") == sent
+
+
+@pytest.mark.parametrize("key_mode", [False, True])
+def test_issue_list_filters(logged_in, runner, key_mode):
+    prefix = "/v1" if key_mode else "/api"
+    with respx.mock as mock:
+        route = mock.get(f"{BASE}{prefix}/issues").respond(200, json=[])
+        result = runner.invoke(app, ["--json", *(["--api-key", "k"] if key_mode else []), "issues", "list",
+                                     "--agent-id", "a1,a2", "--source", "audit", "--assignee", "me",
+                                     "--archived", "all", "--search", "timeout"])
+    assert result.exit_code == 0, result.output
+    params = route.calls[0].request.url.params
+    assert params["agent"] == "a1,a2" and params["source"] == "audit"
+    assert params["assignee"] == "me" and params["archived"] == "all" and params["q"] == "timeout"
+    assert json.loads(result.stdout) == {"issues": []}
+
+
+def test_issue_list_alert_id_alone_keeps_the_alert_route(logged_in, runner):
+    with respx.mock as mock:
+        route = mock.get(f"{BASE}/api/alerts/al1/issues").respond(200, json=[])
+        result = runner.invoke(app, ["--json", "issues", "list", "--alert-id", "al1"])
+    assert result.exit_code == 0, result.output
+    assert route.called
+
+
+def test_issue_count_and_facets(logged_in, runner):
+    with respx.mock as mock:
+        count = mock.get(f"{BASE}/api/issues/count").respond(200, json={"count": 3})
+        facets = mock.get(f"{BASE}/api/issues/facets").respond(200, json={"agents": [{"value": "a1", "count": 3}]})
+        r1 = runner.invoke(app, ["--json", "issues", "count", "--env", "prod"])
+        r2 = runner.invoke(app, ["--json", "issues", "facets", "--state", "firing"])
+    assert r1.exit_code == 0 and json.loads(r1.stdout) == {"count": 3}
+    assert count.calls[0].request.url.params["environment"] == "prod"
+    assert r2.exit_code == 0, r2.output
+    assert facets.calls[0].request.url.params["state"] == "firing"
+
+
+def test_issue_filter_validation(logged_in, runner):
+    with respx.mock(assert_all_called=False):
+        assert runner.invoke(app, ["issues", "list", "--archived", "maybe"]).exit_code == 2
+        assert runner.invoke(app, ["issues", "list", "--source", "slack"]).exit_code == 2
+
+
+@pytest.mark.parametrize("name,path,v1,body,values", [
+    ("trigger_events", "/api/events/trigger_events", "/v1/events/trigger_events", ["PreToolUse"], ["PreToolUse"]),
+    ("eval_envs", "/api/evaluations/environments", "/v1/evaluations/environments", ["prod"], ["prod"]),
+    ("metric_keys", "/api/evaluations/metric-keys", "/v1/evaluations/metric_keys",
+     [{"name": "latency_ms", "unit": "ms", "count": 4}], ["latency_ms"]),
+])
+@pytest.mark.parametrize("key_mode", [False, True])
+def test_new_list_facets(logged_in, runner, name, path, v1, body, values, key_mode):
+    with respx.mock as mock:
+        mock.get(f"{BASE}{v1 if key_mode else path}").respond(200, json=body)
+        result = runner.invoke(app, ["--json", *(["--api-key", "k"] if key_mode else []), "list", name])
+    assert result.exit_code == 0, result.output
+    assert json.loads(result.stdout) == {"kind": name, "values": values}
+
+
+def test_sessions_import_gzips_and_shares_one_cutoff(logged_in, runner, tmp_path):
+    first, second = tmp_path / "proj" / "s1.jsonl", tmp_path / "proj" / "s2.jsonl.gz"
+    first.parent.mkdir()
+    first.write_text('{"type":"user"}\n')
+    second.write_bytes(gzip.compress(b'{"type":"user"}\n'))
+    with respx.mock as mock:
+        route = mock.post(f"{BASE}/api/events/import").respond(200, json={"sessions": 1})
+        result = runner.invoke(app, ["--json", "sessions", "import", str(first), str(second),
+                                     "--format", "claude", "--env", "dev", "--dry-run"])
+    assert result.exit_code == 0, result.output
+    assert len(route.calls) == 2
+    a, b = (c.request for c in route.calls)
+    assert gzip.decompress(a.content) == b'{"type":"user"}\n'
+    assert gzip.decompress(b.content) == b'{"type":"user"}\n'
+    assert a.headers["content-type"] == "application/octet-stream"
+    assert a.url.params["path"] == "proj/s1.jsonl" and a.url.params["format"] == "claude"
+    assert a.url.params["dry_run"] == "true" and a.url.params["environment"] == "dev"
+    assert a.url.params["existing_before"] == b.url.params["existing_before"]
+    out = json.loads(result.stdout)
+    assert [f["file"] for f in out["files"]] == [str(first), str(second)]
+
+
+def test_sessions_import_bad_format_and_missing_file(logged_in, runner, tmp_path):
+    with respx.mock(assert_all_called=False):
+        assert runner.invoke(app, ["sessions", "import", "x", "--format", "zip"]).exit_code == 2
+        assert runner.invoke(app, ["sessions", "import", str(tmp_path / "none.jsonl")]).exit_code == 2
+
+
+def test_sessions_list_still_works_as_a_group_callback(logged_in, runner):
+    with respx.mock as mock:
+        mock.get(f"{BASE}/api/sessions").respond(200, json={"sessions": [], "next_cursor": None})
+        result = runner.invoke(app, ["--json", "sessions", "--limit", "5"])
+    assert result.exit_code == 0, result.output
+
+
+def test_deploy_backfill_allows_judge(logged_in, runner):
+    with respx.mock as mock:
+        mock.get(f"{BASE}/api/evaluator-definitions").respond(200, json=[])
+        mock.post(f"{BASE}/api/evaluator-definitions").respond(201, json={"id": "d1"})
+        backfill = mock.post(f"{BASE}/api/evaluations/backfill").respond(200, json={"enqueued": 2})
+        result = runner.invoke(app, ["--json", "evaluators", "deploy", "relevance", "--judge",
+                                     "--criteria", "Grade relevance", "--threshold", "0.7", "--backfill", "7d"])
+    assert result.exit_code == 0, result.output
+    assert json.loads(backfill.calls[0].request.content)["eval_key"] == "relevance"
+    assert "model budget" in result.stderr
+
+
+def test_usage_windows_needs_a_key(logged_in, runner):
+    with respx.mock as mock:
+        result = runner.invoke(app, ["--json", "usage", "windows"])
+        assert not mock.calls
+    assert result.exit_code == 2
+
+
+def test_sessions_import_public_format_name_maps_to_the_wire_name(logged_in, runner, tmp_path):
+    f = tmp_path / "events.jsonl"
+    f.write_text('{"event_type":"agent_start"}\n')
+    with respx.mock as mock:
+        route = mock.post(f"{BASE}/api/events/import").respond(200, json={})
+        result = runner.invoke(app, ["--json", "sessions", "import", str(f), "--format", "failproofai"])
+    assert result.exit_code == 0, result.output
+    assert route.calls[0].request.url.params["format"] == "agenteye"

package/.next/standalone/fp-cloud-cli/tests/test_whoami.py

@@ -66,7 +66,7 @@ def test_whoami_human_renders_header_and_panels(logged_in, runner):
     assert "62230791-4811-4f04-b388-ae57bdcb422e" in out
     assert "instance admin" in out
     # permissions panel
-    assert "permissions" in out
+    assert "PERMISSIONS" in out
     assert "dashboards" in out and "read" in out and "write" in out and "delete" in out
     # orgs panel + switch hint to the non-active org
     assert "your orgs" in out

package/.next/standalone/hermes-plugin/__init__.py

@@ -104,6 +104,17 @@ def _profile_name() -> str:
 
 def _profile_settings_path(profile: str) -> str | None:
     """Evidence of the profile running this plugin, not a transcript guess."""
+    try:
+        # The host's own resolver first: it honours a per-task profile
+        # override as well as HERMES_HOME, so each profile of a multi-profile
+        # gateway reports its own config (and roster identity), not the
+        # process-wide one.
+        from hermes_constants import get_hermes_home
+
+        hosted = Path(get_hermes_home()).expanduser() / "config.yaml"
+        return str(hosted.absolute()) if hosted.is_file() else None
+    except Exception:
+        pass
     configured = os.environ.get("HERMES_HOME", "").strip()
     if configured:
         selected = Path(configured).expanduser() / "config.yaml"

package/.next/standalone/hermes-plugin/plugin.yaml

@@ -1,5 +1,5 @@
 name: failproofai
-version: 1.0.12-beta.1
+version: 1.0.12
 description: Real-time FailproofAI policy enforcement for Hermes
 manifest_version: 2
 api_version: 1

package/.next/standalone/package.json

@@ -1,6 +1,6 @@
 {
   "name": "failproofai",
-  "version": "1.0.12-beta.1",
+  "version": "1.0.12",
   "description": "Observability and enforcement for AI agent harnesses. 40 built-in policies hooked into 12 of them — Claude Code, Codex, Cursor, Hermes, OpenClaw and more — blocking the tool call before it runs. Local dashboard included, no account needed.",
   "bin": {
     "failproofai": "./dist/cli.mjs",
@@ -118,22 +118,23 @@
     "postcss": "8.5.26",
     "eslint-plugin-react-hooks": "7.0.1",
     "vite": "8.1.6",
-    "undici": "7.29.1",
+    "undici": "7.30.1",
+    "zod": "4.6.6",
     "brace-expansion": "5.0.12",
     "sharp": "0.35.5",
     "source-map-js": "1.2.2",
     "browserslist": "4.28.8"
   },
   "optionalDependencies": {
-    "@failproofai/failproofaid-linux-x64": "1.0.12-beta.1",
-    "@failproofai/failproofaid-linux-arm64": "1.0.12-beta.1",
-    "@failproofai/failproofaid-darwin-x64": "1.0.12-beta.1",
-    "@failproofai/failproofaid-darwin-arm64": "1.0.12-beta.1"
+    "@failproofai/failproofaid-linux-x64": "1.0.12",
+    "@failproofai/failproofaid-linux-arm64": "1.0.12",
+    "@failproofai/failproofaid-darwin-x64": "1.0.12",
+    "@failproofai/failproofaid-darwin-arm64": "1.0.12"
   },
   "failproofaidBinaries": {
-    "linux-x64": "df15e65b60066248038c8ac6d423bffcb0e38637e8a42fa22250afa111fd46ac",
-    "linux-arm64": "59d7ae758c8981c5e0a89eb64be18ebd278213525dd4447891a51be6bb018ed9",
-    "darwin-x64": "b0d3d39abaa02ed0a14dddc68277975a7a23e200830b7edbffdedde3af81ce73",
-    "darwin-arm64": "a6d343b5926e84c0eb437af307f85b69b3128f94c6e34acbf9d0b13a7c7e9d63"
+    "linux-x64": "3b379ce2b03dd8a74393daed2c3120309f591fe6433e0c939d70c7b20da55be7",
+    "linux-arm64": "bd77d7fee68dc2723fb57cc9c7fcf957b7e939c797e61b099c412263880aa16e",
+    "darwin-x64": "fce3faab9406b1c2267955f0e07d04ef7a2b7a0b007e12ecd144fcfef563ebc8",
+    "darwin-arm64": "765732f8f64aa3d3dbd98e93b1aba71fc02956f65baeb73b08447b37fac3a17b"
   }
 }

package/.next/standalone/sdk/python/uv.lock

@@ -1,5 +1,5 @@
 version = 1
-revision = 3
+revision = 5
 requires-python = ">=3.10"
 resolution-markers = [
     "python_full_version >= '3.14' and sys_platform == 'win32'",
@@ -1091,11 +1091,33 @@ provides-extras = ["dev", "langchain", "langgraph", "crewai", "llamaindex", "pyd
 
 [[package]]
 name = "filelock"
-version = "3.32.3"
+version = "4.1.0"
+source = { registry = "https://pypi.org/simple" }
+resolution-markers = [
+    "python_full_version < '3.11' and sys_platform == 'win32'",
+    "python_full_version < '3.11' and sys_platform != 'win32'",
+]
+sdist = { url = "https://files.pythonhosted.org/packages/4c/58/6fd434bec86eff7c38a3168454cb132b762b2bea9b3ac094101a2f7bc32a/filelock-4.1.0.tar.gz", hash = "sha256:ad7f724afef953e731b1cc39bcd3a09166d72ed7fcdf29e6e88b1c3235c6715d", size = 561277, upload-time = "2026-10-09T19:57:20.34Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/ee/86/032133892a5de43b5a98200b01aadcad68cc255e274a762f08b8a76d2912/filelock-4.1.0-py3-none-any.whl", hash = "sha256:2ce9818e3e2d8f284c1a964414447ef148d42a5fd5e2a477a7118e574b293ec1", size = 133003, upload-time = "2026-10-09T19:57:18.716Z" },
+]
+
+[[package]]
+name = "filelock"
+version = "4.2.4"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/7d/64/a02e6765de08964ed371eca577870593245afc9dfac16d037de7c10d18e6/filelock-3.32.3.tar.gz", hash = "sha256:0ffa185a3540854c95caa7fa76b76cb219d907415e2c5dc9af25fd970563487f", size = 218135, upload-time = "2026-08-13T16:00:05.577Z" }
+resolution-markers = [
+    "python_full_version >= '3.14' and sys_platform == 'win32'",
+    "python_full_version >= '3.14' and sys_platform != 'win32'",
+    "python_full_version == '3.13.*'",
+    "python_full_version == '3.12.*' and sys_platform == 'win32'",
+    "python_full_version == '3.12.*' and sys_platform != 'win32'",
+    "python_full_version == '3.11.*' and sys_platform == 'win32'",
+    "python_full_version == '3.11.*' and sys_platform != 'win32'",
+]
+sdist = { url = "https://files.pythonhosted.org/packages/d8/28/3893324bcadf8f89b1616637468d766af3f625008c0735bf704fdfa86617/filelock-4.2.4.tar.gz", hash = "sha256:c6d8663535e1650ee430507b4ffbac7b464dd64b7647ea0112a9f406499d163d", size = 652444, upload-time = "2026-10-19T01:57:51.001Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/a7/8e/50f46a9c0ce8d2861a394c1347caae037ea0431d2f67d7feb151cbc4649a/filelock-3.32.3-py3-none-any.whl", hash = "sha256:7f0ca4bcc0e181c60dbbd8aa9ab5b120ebb99e4e064e83636340056f833a1f09", size = 98901, upload-time = "2026-08-13T16:00:03.974Z" },
+    { url = "https://files.pythonhosted.org/packages/e0/1b/5e74289a1878e666bb4b65fe5805a04cafcac2aa5f519ef1ead67ef91c56/filelock-4.2.4-py3-none-any.whl", hash = "sha256:301241438c57936ef93f4243a94f2559022268bfcf9b375a9a72eb0f8db8358e", size = 162475, upload-time = "2026-10-19T01:57:49.419Z" },
 ]
 
 [[package]]
@@ -1679,7 +1701,8 @@ version = "1.28.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "click" },
-    { name = "filelock" },
+    { name = "filelock", version = "4.1.0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
+    { name = "filelock", version = "4.2.4", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
     { name = "fsspec" },
     { name = "hf-xet", marker = "platform_machine == 'AMD64' or platform_machine == 'aarch64' or platform_machine == 'amd64' or platform_machine == 'arm64' or platform_machine == 'x86_64'" },
     { name = "httpx" },
@@ -4967,11 +4990,11 @@ wheels = [
 
 [[package]]
 name = "setuptools"
-version = "84.0.0"
+version = "84.2.1"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/6d/44/f5da03a8ef95d369145c5bb53050e7877c9f3d312e128605fd9504829143/setuptools-84.0.0.tar.gz", hash = "sha256:f4695c21257f0d9b537ec2692c941d02ee143b7cc1276941349a546573b2ef73", size = 1168449, upload-time = "2026-08-08T18:27:58.365Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/70/b3/c19426f783614d625699a4b247faf628ea9ab1db228674491767b417f9e1/setuptools-84.2.1.tar.gz", hash = "sha256:3ff8b3b5492a0aa66960f4419365a86b84e013070d4313e4d4a6894f5d64d6bc", size = 1230289, upload-time = "2026-10-20T19:01:37.742Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/95/9c/c510029fc6ef33a6275cd2c5d3cecd6613dfd6aa401d57c54f1c18852ccf/setuptools-84.0.0-py3-none-any.whl", hash = "sha256:51a52592b3b99e102b609654876bd65f19f999935166d1352678931132b0c670", size = 818216, upload-time = "2026-08-08T18:27:56.719Z" },
+    { url = "https://files.pythonhosted.org/packages/ef/55/6dc6d6bb953bdfc52519a0059592c52b6e3b8ad51cf8256dffceb55b0610/setuptools-84.2.1-py3-none-any.whl", hash = "sha256:ca140b45d5325a9478f880e351484f62fef15f336462e449ca0c03c873d589c2", size = 876263, upload-time = "2026-10-20T19:01:35.678Z" },
 ]
 
 [[package]]